Most people who install OpenClaw get it working, connect a channel, and stop there. That’s fine — it’s useful even at that level. But the platform has a depth that’s easy to miss, and the features below are where it gets genuinely powerful.
If you’ve had OpenClaw running for a while and haven’t explored these, set aside 30 minutes this weekend. Each one changes how you use it.
—
1. The Skills System
OpenClaw agents can be extended with skills — purpose-built capability packs that give your agent new tools. Skills live in your agent workspace under skills/ and each has a SKILL.md that the agent reads when the task matches.
Skills available via ClawHub include things like:
- Web search (Brave API)
- Whisper audio transcription
- Image generation
- Weather lookups
- Custom integrations
Install from the community hub:
# Using the clawhub skill
# Ask your agent: "install the weather skill from clawhub"
Or build your own — a skill is just a folder with a SKILL.md and any supporting scripts. If you find yourself giving your agent the same multi-step instruction repeatedly, it probably belongs in a skill.
—
2. Persistent, File-Based Memory
Unlike hosted AI assistants where “memory” is opaque and limited, OpenClaw’s memory system is yours and readable.
Your agent maintains:
MEMORY.md — long-term curated memory (big decisions, preferences, important context)memory/YYYY-MM-DD.md — daily logs of what happenedUSER.md — everything the agent knows about youSOUL.md — the agent’s personality and behavioural rules
You can read these files, edit them, back them up, and migrate them to a new machine. Memory is just files. That’s a feature.
To update your agent’s long-term memory, just tell it: “Remember that I prefer concise responses and my timezone is CET.” It writes it to MEMORY.md.
—
3. Proactive Heartbeats
Most AI assistants are reactive — they wait for you to ask. OpenClaw can be proactive.
The heartbeat system sends your agent a periodic poll (every 30 minutes by default) to check if anything needs attention. Your agent can be configured to:
- Check for important emails and ping you about them
- Surface upcoming calendar events before they sneak up on you
- Monitor a project status and alert you to blockers
- Do background tasks and update you with a summary
Configure what your agent checks in HEARTBEAT.md in the workspace. Keep it short — the agent reads and acts on it every heartbeat cycle.
This is one of the most underused features and one of the most transformative. You go from “AI I have to remember to ask” to “AI that reminds me of things.”
—
4. Cron Jobs for Scheduled Agent Tasks
Beyond heartbeats, you can schedule precise, isolated agent tasks with cron:
# Ask your agent to set this up:
"Every Monday at 9am, summarise what's on my calendar for the week and send it to me"
Or set it up directly in config. Cron jobs can:
- Run at exact times (not just periodic intervals)
- Use a different model or thinking level than your main agent
- Deliver results directly to a channel without bothering your main session
- Trigger one-shot reminders (“remind me about X in 20 minutes”)
This is the difference between an AI assistant and an AI that actually does things while you’re not looking.
—
5. Multi-Agent Routing
One Gateway, multiple isolated agents. Each agent has its own:
- Workspace (files, memory, persona)
- Channel bindings (different WhatsApp/Telegram accounts)
- Model configuration
- Skills
Why this matters: You can have a personal agent on your personal WhatsApp and a work agent on your work Discord — completely isolated, different personas, different permissions — running from the same machine.
Add a new agent:
openclaw agents add work
openclaw agents list --bindings
—
6. Connecting Every Chat App At Once
Most people connect one channel and forget the rest. OpenClaw supports:
- WhatsApp
- Telegram
- Discord
- iMessage (macOS)
- Signal
- Slack
- Google Chat
- Microsoft Teams
- Mattermost
- IRC, Nostr, and more via plugins
All channels share the same Gateway. Your agent is the same agent everywhere — same memory, same context. You can start a conversation on Telegram and continue it on Discord.
—
7. Local Models via Ollama
OpenClaw works with any OpenAI-compatible API, which means you can point it at a local Ollama instance for completely offline, zero-cost, fully private inference.
In your config:
{
"agents": {
"defaults": {
"model": "ollama/llama3.2"
}
}
}
Run everything locally on a reasonably specced machine. Great for privacy-sensitive tasks or when you want to avoid API costs for high-volume work.
—
8. Mobile Nodes with Canvas
OpenClaw can pair with iOS and Android phones as nodes — not just as messaging endpoints, but as connected devices that can:
- Share camera access with your agent
- Display a Canvas (a rendered UI surface your agent can push to your phone)
- Report location
- Run agent-triggered actions
The pairing is simple:
openclaw onboard # follow the nodes section
This is genuinely underexplored. The ability for your agent to push a rendered UI to your phone — triggered by a conversation in Telegram — is a different category of interaction.
—
9. The Web Control UI
Beyond messaging, OpenClaw ships with a full browser-based Control UI at http://127.0.0.1:18789/. From here you can:
- Chat with your agent
- Browse session history
- View and edit config with a form UI (no JSON editing required)
- Monitor gateway status
- Manage channels and pairings
openclaw dashboard # opens it in your browser
If you’ve only ever used OpenClaw through a messaging app, spend five minutes in the Control UI. The config form editor alone saves a lot of manual JSON work.
—
10. Config Hot Reload
This one is small but great: OpenClaw watches your config file (~/.openclaw/openclaw.json) and reloads changes automatically. No restart required.
Change your model, adjust an allowlist, tweak a channel setting — save the file, it takes effect. This makes iterating on your configuration dramatically less painful, especially when you’re dialling in agent behaviour or channel permissions.
—
Where to Go From Here
If you’re just getting started with OpenClaw, pick one feature from this list to explore this week. The heartbeat system and memory files tend to deliver the biggest “aha” moments for new users.
If you’re already using several of these — what’s missing from the list? What’s in your OpenClaw setup that you don’t see enough people talking about? Drop it in the comments. 👇
